1 // SPDX-License-Identifier: GPL-2.0-only
5 * Copyright (C) 1991, 1992 Linus Torvalds
9 #include <linux/slab.h>
10 #include <linux/sched/autogroup.h>
11 #include <linux/sched/mm.h>
12 #include <linux/sched/stat.h>
13 #include <linux/sched/task.h>
14 #include <linux/sched/task_stack.h>
15 #include <linux/sched/cputime.h>
16 #include <linux/interrupt.h>
17 #include <linux/module.h>
18 #include <linux/capability.h>
19 #include <linux/completion.h>
20 #include <linux/personality.h>
21 #include <linux/tty.h>
22 #include <linux/iocontext.h>
23 #include <linux/key.h>
24 #include <linux/cpu.h>
25 #include <linux/acct.h>
26 #include <linux/tsacct_kern.h>
27 #include <linux/file.h>
28 #include <linux/fdtable.h>
29 #include <linux/freezer.h>
30 #include <linux/binfmts.h>
31 #include <linux/nsproxy.h>
32 #include <linux/pid_namespace.h>
33 #include <linux/ptrace.h>
34 #include <linux/profile.h>
35 #include <linux/mount.h>
36 #include <linux/proc_fs.h>
37 #include <linux/kthread.h>
38 #include <linux/mempolicy.h>
39 #include <linux/taskstats_kern.h>
40 #include <linux/delayacct.h>
41 #include <linux/cgroup.h>
42 #include <linux/syscalls.h>
43 #include <linux/signal.h>
44 #include <linux/posix-timers.h>
45 #include <linux/cn_proc.h>
46 #include <linux/mutex.h>
47 #include <linux/futex.h>
48 #include <linux/pipe_fs_i.h>
49 #include <linux/audit.h> /* for audit_free() */
50 #include <linux/resource.h>
51 #include <linux/blkdev.h>
52 #include <linux/task_io_accounting_ops.h>
53 #include <linux/tracehook.h>
54 #include <linux/fs_struct.h>
55 #include <linux/init_task.h>
56 #include <linux/perf_event.h>
57 #include <trace/events/sched.h>
58 #include <linux/hw_breakpoint.h>
59 #include <linux/oom.h>
60 #include <linux/writeback.h>
61 #include <linux/shm.h>
62 #include <linux/kcov.h>
63 #include <linux/random.h>
64 #include <linux/rcuwait.h>
65 #include <linux/compat.h>
67 #include <linux/uaccess.h>
68 #include <asm/unistd.h>
69 #include <asm/pgtable.h>
70 #include <asm/mmu_context.h>
72 static void __unhash_process(struct task_struct
*p
, bool group_dead
)
75 detach_pid(p
, PIDTYPE_PID
);
77 detach_pid(p
, PIDTYPE_TGID
);
78 detach_pid(p
, PIDTYPE_PGID
);
79 detach_pid(p
, PIDTYPE_SID
);
81 list_del_rcu(&p
->tasks
);
82 list_del_init(&p
->sibling
);
83 __this_cpu_dec(process_counts
);
85 list_del_rcu(&p
->thread_group
);
86 list_del_rcu(&p
->thread_node
);
90 * This function expects the tasklist_lock write-locked.
92 static void __exit_signal(struct task_struct
*tsk
)
94 struct signal_struct
*sig
= tsk
->signal
;
95 bool group_dead
= thread_group_leader(tsk
);
96 struct sighand_struct
*sighand
;
97 struct tty_struct
*uninitialized_var(tty
);
100 sighand
= rcu_dereference_check(tsk
->sighand
,
101 lockdep_tasklist_lock_is_held());
102 spin_lock(&sighand
->siglock
);
104 #ifdef CONFIG_POSIX_TIMERS
105 posix_cpu_timers_exit(tsk
);
107 posix_cpu_timers_exit_group(tsk
);
115 * If there is any task waiting for the group exit
118 if (sig
->notify_count
> 0 && !--sig
->notify_count
)
119 wake_up_process(sig
->group_exit_task
);
121 if (tsk
== sig
->curr_target
)
122 sig
->curr_target
= next_thread(tsk
);
125 add_device_randomness((const void*) &tsk
->se
.sum_exec_runtime
,
126 sizeof(unsigned long long));
129 * Accumulate here the counters for all threads as they die. We could
130 * skip the group leader because it is the last user of signal_struct,
131 * but we want to avoid the race with thread_group_cputime() which can
132 * see the empty ->thread_head list.
134 task_cputime(tsk
, &utime
, &stime
);
135 write_seqlock(&sig
->stats_lock
);
138 sig
->gtime
+= task_gtime(tsk
);
139 sig
->min_flt
+= tsk
->min_flt
;
140 sig
->maj_flt
+= tsk
->maj_flt
;
141 sig
->nvcsw
+= tsk
->nvcsw
;
142 sig
->nivcsw
+= tsk
->nivcsw
;
143 sig
->inblock
+= task_io_get_inblock(tsk
);
144 sig
->oublock
+= task_io_get_oublock(tsk
);
145 task_io_accounting_add(&sig
->ioac
, &tsk
->ioac
);
146 sig
->sum_sched_runtime
+= tsk
->se
.sum_exec_runtime
;
148 __unhash_process(tsk
, group_dead
);
149 write_sequnlock(&sig
->stats_lock
);
152 * Do this under ->siglock, we can race with another thread
153 * doing sigqueue_free() if we have SIGQUEUE_PREALLOC signals.
155 flush_sigqueue(&tsk
->pending
);
157 spin_unlock(&sighand
->siglock
);
159 __cleanup_sighand(sighand
);
160 clear_tsk_thread_flag(tsk
, TIF_SIGPENDING
);
162 flush_sigqueue(&sig
->shared_pending
);
167 static void delayed_put_task_struct(struct rcu_head
*rhp
)
169 struct task_struct
*tsk
= container_of(rhp
, struct task_struct
, rcu
);
171 perf_event_delayed_put(tsk
);
172 trace_sched_process_free(tsk
);
173 put_task_struct(tsk
);
176 void put_task_struct_rcu_user(struct task_struct
*task
)
178 if (refcount_dec_and_test(&task
->rcu_users
))
179 call_rcu(&task
->rcu
, delayed_put_task_struct
);
182 void release_task(struct task_struct
*p
)
184 struct task_struct
*leader
;
185 struct pid
*thread_pid
;
188 /* don't need to get the RCU readlock here - the process is dead and
189 * can't be modifying its own credentials. But shut RCU-lockdep up */
191 atomic_dec(&__task_cred(p
)->user
->processes
);
196 write_lock_irq(&tasklist_lock
);
197 ptrace_release_task(p
);
198 thread_pid
= get_pid(p
->thread_pid
);
202 * If we are the last non-leader member of the thread
203 * group, and the leader is zombie, then notify the
204 * group leader's parent process. (if it wants notification.)
207 leader
= p
->group_leader
;
208 if (leader
!= p
&& thread_group_empty(leader
)
209 && leader
->exit_state
== EXIT_ZOMBIE
) {
211 * If we were the last child thread and the leader has
212 * exited already, and the leader's parent ignores SIGCHLD,
213 * then we are the one who should release the leader.
215 zap_leader
= do_notify_parent(leader
, leader
->exit_signal
);
217 leader
->exit_state
= EXIT_DEAD
;
220 write_unlock_irq(&tasklist_lock
);
221 proc_flush_pid(thread_pid
);
223 put_task_struct_rcu_user(p
);
226 if (unlikely(zap_leader
))
230 void rcuwait_wake_up(struct rcuwait
*w
)
232 struct task_struct
*task
;
237 * Order condition vs @task, such that everything prior to the load
238 * of @task is visible. This is the condition as to why the user called
239 * rcuwait_trywake() in the first place. Pairs with set_current_state()
240 * barrier (A) in rcuwait_wait_event().
243 * [S] tsk = current [S] cond = true
249 task
= rcu_dereference(w
->task
);
251 wake_up_process(task
);
254 EXPORT_SYMBOL_GPL(rcuwait_wake_up
);
257 * Determine if a process group is "orphaned", according to the POSIX
258 * definition in 2.2.2.52. Orphaned process groups are not to be affected
259 * by terminal-generated stop signals. Newly orphaned process groups are
260 * to receive a SIGHUP and a SIGCONT.
262 * "I ask you, have you ever known what it is to be an orphan?"
264 static int will_become_orphaned_pgrp(struct pid
*pgrp
,
265 struct task_struct
*ignored_task
)
267 struct task_struct
*p
;
269 do_each_pid_task(pgrp
, PIDTYPE_PGID
, p
) {
270 if ((p
== ignored_task
) ||
271 (p
->exit_state
&& thread_group_empty(p
)) ||
272 is_global_init(p
->real_parent
))
275 if (task_pgrp(p
->real_parent
) != pgrp
&&
276 task_session(p
->real_parent
) == task_session(p
))
278 } while_each_pid_task(pgrp
, PIDTYPE_PGID
, p
);
283 int is_current_pgrp_orphaned(void)
287 read_lock(&tasklist_lock
);
288 retval
= will_become_orphaned_pgrp(task_pgrp(current
), NULL
);
289 read_unlock(&tasklist_lock
);
294 static bool has_stopped_jobs(struct pid
*pgrp
)
296 struct task_struct
*p
;
298 do_each_pid_task(pgrp
, PIDTYPE_PGID
, p
) {
299 if (p
->signal
->flags
& SIGNAL_STOP_STOPPED
)
301 } while_each_pid_task(pgrp
, PIDTYPE_PGID
, p
);
307 * Check to see if any process groups have become orphaned as
308 * a result of our exiting, and if they have any stopped jobs,
309 * send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
312 kill_orphaned_pgrp(struct task_struct
*tsk
, struct task_struct
*parent
)
314 struct pid
*pgrp
= task_pgrp(tsk
);
315 struct task_struct
*ignored_task
= tsk
;
318 /* exit: our father is in a different pgrp than
319 * we are and we were the only connection outside.
321 parent
= tsk
->real_parent
;
323 /* reparent: our child is in a different pgrp than
324 * we are, and it was the only connection outside.
328 if (task_pgrp(parent
) != pgrp
&&
329 task_session(parent
) == task_session(tsk
) &&
330 will_become_orphaned_pgrp(pgrp
, ignored_task
) &&
331 has_stopped_jobs(pgrp
)) {
332 __kill_pgrp_info(SIGHUP
, SEND_SIG_PRIV
, pgrp
);
333 __kill_pgrp_info(SIGCONT
, SEND_SIG_PRIV
, pgrp
);
339 * A task is exiting. If it owned this mm, find a new owner for the mm.
341 void mm_update_next_owner(struct mm_struct
*mm
)
343 struct task_struct
*c
, *g
, *p
= current
;
347 * If the exiting or execing task is not the owner, it's
348 * someone else's problem.
353 * The current owner is exiting/execing and there are no other
354 * candidates. Do not leave the mm pointing to a possibly
355 * freed task structure.
357 if (atomic_read(&mm
->mm_users
) <= 1) {
358 WRITE_ONCE(mm
->owner
, NULL
);
362 read_lock(&tasklist_lock
);
364 * Search in the children
366 list_for_each_entry(c
, &p
->children
, sibling
) {
368 goto assign_new_owner
;
372 * Search in the siblings
374 list_for_each_entry(c
, &p
->real_parent
->children
, sibling
) {
376 goto assign_new_owner
;
380 * Search through everything else, we should not get here often.
382 for_each_process(g
) {
383 if (g
->flags
& PF_KTHREAD
)
385 for_each_thread(g
, c
) {
387 goto assign_new_owner
;
392 read_unlock(&tasklist_lock
);
394 * We found no owner yet mm_users > 1: this implies that we are
395 * most likely racing with swapoff (try_to_unuse()) or /proc or
396 * ptrace or page migration (get_task_mm()). Mark owner as NULL.
398 WRITE_ONCE(mm
->owner
, NULL
);
405 * The task_lock protects c->mm from changing.
406 * We always want mm->owner->mm == mm
410 * Delay read_unlock() till we have the task_lock()
411 * to ensure that c does not slip away underneath us
413 read_unlock(&tasklist_lock
);
419 WRITE_ONCE(mm
->owner
, c
);
423 #endif /* CONFIG_MEMCG */
426 * Turn us into a lazy TLB process if we
429 static void exit_mm(void)
431 struct mm_struct
*mm
= current
->mm
;
432 struct core_state
*core_state
;
434 exit_mm_release(current
, mm
);
439 * Serialize with any possible pending coredump.
440 * We must hold mmap_sem around checking core_state
441 * and clearing tsk->mm. The core-inducing thread
442 * will increment ->nr_threads for each thread in the
443 * group with ->mm != NULL.
445 down_read(&mm
->mmap_sem
);
446 core_state
= mm
->core_state
;
448 struct core_thread self
;
450 up_read(&mm
->mmap_sem
);
453 self
.next
= xchg(&core_state
->dumper
.next
, &self
);
455 * Implies mb(), the result of xchg() must be visible
456 * to core_state->dumper.
458 if (atomic_dec_and_test(&core_state
->nr_threads
))
459 complete(&core_state
->startup
);
462 set_current_state(TASK_UNINTERRUPTIBLE
);
463 if (!self
.task
) /* see coredump_finish() */
465 freezable_schedule();
467 __set_current_state(TASK_RUNNING
);
468 down_read(&mm
->mmap_sem
);
471 BUG_ON(mm
!= current
->active_mm
);
472 /* more a memory barrier than a real lock */
475 up_read(&mm
->mmap_sem
);
476 enter_lazy_tlb(mm
, current
);
477 task_unlock(current
);
478 mm_update_next_owner(mm
);
480 if (test_thread_flag(TIF_MEMDIE
))
484 static struct task_struct
*find_alive_thread(struct task_struct
*p
)
486 struct task_struct
*t
;
488 for_each_thread(p
, t
) {
489 if (!(t
->flags
& PF_EXITING
))
495 static struct task_struct
*find_child_reaper(struct task_struct
*father
,
496 struct list_head
*dead
)
497 __releases(&tasklist_lock
)
498 __acquires(&tasklist_lock
)
500 struct pid_namespace
*pid_ns
= task_active_pid_ns(father
);
501 struct task_struct
*reaper
= pid_ns
->child_reaper
;
502 struct task_struct
*p
, *n
;
504 if (likely(reaper
!= father
))
507 reaper
= find_alive_thread(father
);
509 pid_ns
->child_reaper
= reaper
;
513 write_unlock_irq(&tasklist_lock
);
515 list_for_each_entry_safe(p
, n
, dead
, ptrace_entry
) {
516 list_del_init(&p
->ptrace_entry
);
520 zap_pid_ns_processes(pid_ns
);
521 write_lock_irq(&tasklist_lock
);
527 * When we die, we re-parent all our children, and try to:
528 * 1. give them to another thread in our thread group, if such a member exists
529 * 2. give it to the first ancestor process which prctl'd itself as a
530 * child_subreaper for its children (like a service manager)
531 * 3. give it to the init process (PID 1) in our pid namespace
533 static struct task_struct
*find_new_reaper(struct task_struct
*father
,
534 struct task_struct
*child_reaper
)
536 struct task_struct
*thread
, *reaper
;
538 thread
= find_alive_thread(father
);
542 if (father
->signal
->has_child_subreaper
) {
543 unsigned int ns_level
= task_pid(father
)->level
;
545 * Find the first ->is_child_subreaper ancestor in our pid_ns.
546 * We can't check reaper != child_reaper to ensure we do not
547 * cross the namespaces, the exiting parent could be injected
548 * by setns() + fork().
549 * We check pid->level, this is slightly more efficient than
550 * task_active_pid_ns(reaper) != task_active_pid_ns(father).
552 for (reaper
= father
->real_parent
;
553 task_pid(reaper
)->level
== ns_level
;
554 reaper
= reaper
->real_parent
) {
555 if (reaper
== &init_task
)
557 if (!reaper
->signal
->is_child_subreaper
)
559 thread
= find_alive_thread(reaper
);
569 * Any that need to be release_task'd are put on the @dead list.
571 static void reparent_leader(struct task_struct
*father
, struct task_struct
*p
,
572 struct list_head
*dead
)
574 if (unlikely(p
->exit_state
== EXIT_DEAD
))
577 /* We don't want people slaying init. */
578 p
->exit_signal
= SIGCHLD
;
580 /* If it has exited notify the new parent about this child's death. */
582 p
->exit_state
== EXIT_ZOMBIE
&& thread_group_empty(p
)) {
583 if (do_notify_parent(p
, p
->exit_signal
)) {
584 p
->exit_state
= EXIT_DEAD
;
585 list_add(&p
->ptrace_entry
, dead
);
589 kill_orphaned_pgrp(p
, father
);
593 * This does two things:
595 * A. Make init inherit all the child processes
596 * B. Check to see if any process groups have become orphaned
597 * as a result of our exiting, and if they have any stopped
598 * jobs, send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
600 static void forget_original_parent(struct task_struct
*father
,
601 struct list_head
*dead
)
603 struct task_struct
*p
, *t
, *reaper
;
605 if (unlikely(!list_empty(&father
->ptraced
)))
606 exit_ptrace(father
, dead
);
608 /* Can drop and reacquire tasklist_lock */
609 reaper
= find_child_reaper(father
, dead
);
610 if (list_empty(&father
->children
))
613 reaper
= find_new_reaper(father
, reaper
);
614 list_for_each_entry(p
, &father
->children
, sibling
) {
615 for_each_thread(p
, t
) {
616 RCU_INIT_POINTER(t
->real_parent
, reaper
);
617 BUG_ON((!t
->ptrace
) != (rcu_access_pointer(t
->parent
) == father
));
618 if (likely(!t
->ptrace
))
619 t
->parent
= t
->real_parent
;
620 if (t
->pdeath_signal
)
621 group_send_sig_info(t
->pdeath_signal
,
626 * If this is a threaded reparent there is no need to
627 * notify anyone anything has happened.
629 if (!same_thread_group(reaper
, father
))
630 reparent_leader(father
, p
, dead
);
632 list_splice_tail_init(&father
->children
, &reaper
->children
);
636 * Send signals to all our closest relatives so that they know
637 * to properly mourn us..
639 static void exit_notify(struct task_struct
*tsk
, int group_dead
)
642 struct task_struct
*p
, *n
;
645 write_lock_irq(&tasklist_lock
);
646 forget_original_parent(tsk
, &dead
);
649 kill_orphaned_pgrp(tsk
->group_leader
, NULL
);
651 tsk
->exit_state
= EXIT_ZOMBIE
;
652 if (unlikely(tsk
->ptrace
)) {
653 int sig
= thread_group_leader(tsk
) &&
654 thread_group_empty(tsk
) &&
655 !ptrace_reparented(tsk
) ?
656 tsk
->exit_signal
: SIGCHLD
;
657 autoreap
= do_notify_parent(tsk
, sig
);
658 } else if (thread_group_leader(tsk
)) {
659 autoreap
= thread_group_empty(tsk
) &&
660 do_notify_parent(tsk
, tsk
->exit_signal
);
666 tsk
->exit_state
= EXIT_DEAD
;
667 list_add(&tsk
->ptrace_entry
, &dead
);
670 /* mt-exec, de_thread() is waiting for group leader */
671 if (unlikely(tsk
->signal
->notify_count
< 0))
672 wake_up_process(tsk
->signal
->group_exit_task
);
673 write_unlock_irq(&tasklist_lock
);
675 list_for_each_entry_safe(p
, n
, &dead
, ptrace_entry
) {
676 list_del_init(&p
->ptrace_entry
);
681 #ifdef CONFIG_DEBUG_STACK_USAGE
682 static void check_stack_usage(void)
684 static DEFINE_SPINLOCK(low_water_lock
);
685 static int lowest_to_date
= THREAD_SIZE
;
688 free
= stack_not_used(current
);
690 if (free
>= lowest_to_date
)
693 spin_lock(&low_water_lock
);
694 if (free
< lowest_to_date
) {
695 pr_info("%s (%d) used greatest stack depth: %lu bytes left\n",
696 current
->comm
, task_pid_nr(current
), free
);
697 lowest_to_date
= free
;
699 spin_unlock(&low_water_lock
);
702 static inline void check_stack_usage(void) {}
705 void __noreturn
do_exit(long code
)
707 struct task_struct
*tsk
= current
;
710 profile_task_exit(tsk
);
713 WARN_ON(blk_needs_flush_plug(tsk
));
715 if (unlikely(in_interrupt()))
716 panic("Aiee, killing interrupt handler!");
717 if (unlikely(!tsk
->pid
))
718 panic("Attempted to kill the idle task!");
721 * If do_exit is called because this processes oopsed, it's possible
722 * that get_fs() was left as KERNEL_DS, so reset it to USER_DS before
723 * continuing. Amongst other possible reasons, this is to prevent
724 * mm_release()->clear_child_tid() from writing to a user-controlled
729 ptrace_event(PTRACE_EVENT_EXIT
, code
);
731 validate_creds_for_do_exit(tsk
);
734 * We're taking recursive faults here in do_exit. Safest is to just
735 * leave this task alone and wait for reboot.
737 if (unlikely(tsk
->flags
& PF_EXITING
)) {
738 pr_alert("Fixing recursive fault but reboot is needed!\n");
739 futex_exit_recursive(tsk
);
740 set_current_state(TASK_UNINTERRUPTIBLE
);
744 exit_signals(tsk
); /* sets PF_EXITING */
746 if (unlikely(in_atomic())) {
747 pr_info("note: %s[%d] exited with preempt_count %d\n",
748 current
->comm
, task_pid_nr(current
),
750 preempt_count_set(PREEMPT_ENABLED
);
753 /* sync mm's RSS info before statistics gathering */
755 sync_mm_rss(tsk
->mm
);
756 acct_update_integrals(tsk
);
757 group_dead
= atomic_dec_and_test(&tsk
->signal
->live
);
760 * If the last thread of global init has exited, panic
761 * immediately to get a useable coredump.
763 if (unlikely(is_global_init(tsk
)))
764 panic("Attempted to kill init! exitcode=0x%08x\n",
765 tsk
->signal
->group_exit_code
?: (int)code
);
767 #ifdef CONFIG_POSIX_TIMERS
768 hrtimer_cancel(&tsk
->signal
->real_timer
);
769 exit_itimers(tsk
->signal
);
772 setmax_mm_hiwater_rss(&tsk
->signal
->maxrss
, tsk
->mm
);
774 acct_collect(code
, group_dead
);
779 tsk
->exit_code
= code
;
780 taskstats_exit(tsk
, group_dead
);
786 trace_sched_process_exit(tsk
);
793 disassociate_ctty(1);
794 exit_task_namespaces(tsk
);
800 * Flush inherited counters to the parent - before the parent
801 * gets woken up by child-exit notifications.
803 * because of cgroup mode, must be called before cgroup_exit()
805 perf_event_exit_task(tsk
);
807 sched_autogroup_exit_task(tsk
);
811 * FIXME: do that only when needed, using sched_exit tracepoint
813 flush_ptrace_hw_breakpoint(tsk
);
815 exit_tasks_rcu_start();
816 exit_notify(tsk
, group_dead
);
817 proc_exit_connector(tsk
);
818 mpol_put_task_policy(tsk
);
820 if (unlikely(current
->pi_state_cache
))
821 kfree(current
->pi_state_cache
);
824 * Make sure we are holding no locks:
826 debug_check_no_locks_held();
829 exit_io_context(tsk
);
831 if (tsk
->splice_pipe
)
832 free_pipe_info(tsk
->splice_pipe
);
834 if (tsk
->task_frag
.page
)
835 put_page(tsk
->task_frag
.page
);
837 validate_creds_for_do_exit(tsk
);
842 __this_cpu_add(dirty_throttle_leaks
, tsk
->nr_dirtied
);
844 exit_tasks_rcu_finish();
846 lockdep_free_task(tsk
);
849 EXPORT_SYMBOL_GPL(do_exit
);
851 void complete_and_exit(struct completion
*comp
, long code
)
858 EXPORT_SYMBOL(complete_and_exit
);
860 SYSCALL_DEFINE1(exit
, int, error_code
)
862 do_exit((error_code
&0xff)<<8);
866 * Take down every thread in the group. This is called by fatal signals
867 * as well as by sys_exit_group (below).
870 do_group_exit(int exit_code
)
872 struct signal_struct
*sig
= current
->signal
;
874 BUG_ON(exit_code
& 0x80); /* core dumps don't get here */
876 if (signal_group_exit(sig
))
877 exit_code
= sig
->group_exit_code
;
878 else if (!thread_group_empty(current
)) {
879 struct sighand_struct
*const sighand
= current
->sighand
;
881 spin_lock_irq(&sighand
->siglock
);
882 if (signal_group_exit(sig
))
883 /* Another thread got here before we took the lock. */
884 exit_code
= sig
->group_exit_code
;
886 sig
->group_exit_code
= exit_code
;
887 sig
->flags
= SIGNAL_GROUP_EXIT
;
888 zap_other_threads(current
);
890 spin_unlock_irq(&sighand
->siglock
);
898 * this kills every thread in the thread group. Note that any externally
899 * wait4()-ing process will get the correct exit code - even if this
900 * thread is not the thread group leader.
902 SYSCALL_DEFINE1(exit_group
, int, error_code
)
904 do_group_exit((error_code
& 0xff) << 8);
917 enum pid_type wo_type
;
921 struct waitid_info
*wo_info
;
923 struct rusage
*wo_rusage
;
925 wait_queue_entry_t child_wait
;
929 static int eligible_pid(struct wait_opts
*wo
, struct task_struct
*p
)
931 return wo
->wo_type
== PIDTYPE_MAX
||
932 task_pid_type(p
, wo
->wo_type
) == wo
->wo_pid
;
936 eligible_child(struct wait_opts
*wo
, bool ptrace
, struct task_struct
*p
)
938 if (!eligible_pid(wo
, p
))
942 * Wait for all children (clone and not) if __WALL is set or
943 * if it is traced by us.
945 if (ptrace
|| (wo
->wo_flags
& __WALL
))
949 * Otherwise, wait for clone children *only* if __WCLONE is set;
950 * otherwise, wait for non-clone children *only*.
952 * Note: a "clone" child here is one that reports to its parent
953 * using a signal other than SIGCHLD, or a non-leader thread which
954 * we can only see if it is traced by us.
956 if ((p
->exit_signal
!= SIGCHLD
) ^ !!(wo
->wo_flags
& __WCLONE
))
963 * Handle sys_wait4 work for one task in state EXIT_ZOMBIE. We hold
964 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
965 * the lock and this task is uninteresting. If we return nonzero, we have
966 * released the lock and the system call should return.
968 static int wait_task_zombie(struct wait_opts
*wo
, struct task_struct
*p
)
971 pid_t pid
= task_pid_vnr(p
);
972 uid_t uid
= from_kuid_munged(current_user_ns(), task_uid(p
));
973 struct waitid_info
*infop
;
975 if (!likely(wo
->wo_flags
& WEXITED
))
978 if (unlikely(wo
->wo_flags
& WNOWAIT
)) {
979 status
= p
->exit_code
;
981 read_unlock(&tasklist_lock
);
982 sched_annotate_sleep();
984 getrusage(p
, RUSAGE_BOTH
, wo
->wo_rusage
);
989 * Move the task's state to DEAD/TRACE, only one thread can do this.
991 state
= (ptrace_reparented(p
) && thread_group_leader(p
)) ?
992 EXIT_TRACE
: EXIT_DEAD
;
993 if (cmpxchg(&p
->exit_state
, EXIT_ZOMBIE
, state
) != EXIT_ZOMBIE
)
996 * We own this thread, nobody else can reap it.
998 read_unlock(&tasklist_lock
);
999 sched_annotate_sleep();
1002 * Check thread_group_leader() to exclude the traced sub-threads.
1004 if (state
== EXIT_DEAD
&& thread_group_leader(p
)) {
1005 struct signal_struct
*sig
= p
->signal
;
1006 struct signal_struct
*psig
= current
->signal
;
1007 unsigned long maxrss
;
1008 u64 tgutime
, tgstime
;
1011 * The resource counters for the group leader are in its
1012 * own task_struct. Those for dead threads in the group
1013 * are in its signal_struct, as are those for the child
1014 * processes it has previously reaped. All these
1015 * accumulate in the parent's signal_struct c* fields.
1017 * We don't bother to take a lock here to protect these
1018 * p->signal fields because the whole thread group is dead
1019 * and nobody can change them.
1021 * psig->stats_lock also protects us from our sub-theads
1022 * which can reap other children at the same time. Until
1023 * we change k_getrusage()-like users to rely on this lock
1024 * we have to take ->siglock as well.
1026 * We use thread_group_cputime_adjusted() to get times for
1027 * the thread group, which consolidates times for all threads
1028 * in the group including the group leader.
1030 thread_group_cputime_adjusted(p
, &tgutime
, &tgstime
);
1031 spin_lock_irq(¤t
->sighand
->siglock
);
1032 write_seqlock(&psig
->stats_lock
);
1033 psig
->cutime
+= tgutime
+ sig
->cutime
;
1034 psig
->cstime
+= tgstime
+ sig
->cstime
;
1035 psig
->cgtime
+= task_gtime(p
) + sig
->gtime
+ sig
->cgtime
;
1037 p
->min_flt
+ sig
->min_flt
+ sig
->cmin_flt
;
1039 p
->maj_flt
+ sig
->maj_flt
+ sig
->cmaj_flt
;
1041 p
->nvcsw
+ sig
->nvcsw
+ sig
->cnvcsw
;
1043 p
->nivcsw
+ sig
->nivcsw
+ sig
->cnivcsw
;
1045 task_io_get_inblock(p
) +
1046 sig
->inblock
+ sig
->cinblock
;
1048 task_io_get_oublock(p
) +
1049 sig
->oublock
+ sig
->coublock
;
1050 maxrss
= max(sig
->maxrss
, sig
->cmaxrss
);
1051 if (psig
->cmaxrss
< maxrss
)
1052 psig
->cmaxrss
= maxrss
;
1053 task_io_accounting_add(&psig
->ioac
, &p
->ioac
);
1054 task_io_accounting_add(&psig
->ioac
, &sig
->ioac
);
1055 write_sequnlock(&psig
->stats_lock
);
1056 spin_unlock_irq(¤t
->sighand
->siglock
);
1060 getrusage(p
, RUSAGE_BOTH
, wo
->wo_rusage
);
1061 status
= (p
->signal
->flags
& SIGNAL_GROUP_EXIT
)
1062 ? p
->signal
->group_exit_code
: p
->exit_code
;
1063 wo
->wo_stat
= status
;
1065 if (state
== EXIT_TRACE
) {
1066 write_lock_irq(&tasklist_lock
);
1067 /* We dropped tasklist, ptracer could die and untrace */
1070 /* If parent wants a zombie, don't release it now */
1071 state
= EXIT_ZOMBIE
;
1072 if (do_notify_parent(p
, p
->exit_signal
))
1074 p
->exit_state
= state
;
1075 write_unlock_irq(&tasklist_lock
);
1077 if (state
== EXIT_DEAD
)
1081 infop
= wo
->wo_info
;
1083 if ((status
& 0x7f) == 0) {
1084 infop
->cause
= CLD_EXITED
;
1085 infop
->status
= status
>> 8;
1087 infop
->cause
= (status
& 0x80) ? CLD_DUMPED
: CLD_KILLED
;
1088 infop
->status
= status
& 0x7f;
1097 static int *task_stopped_code(struct task_struct
*p
, bool ptrace
)
1100 if (task_is_traced(p
) && !(p
->jobctl
& JOBCTL_LISTENING
))
1101 return &p
->exit_code
;
1103 if (p
->signal
->flags
& SIGNAL_STOP_STOPPED
)
1104 return &p
->signal
->group_exit_code
;
1110 * wait_task_stopped - Wait for %TASK_STOPPED or %TASK_TRACED
1112 * @ptrace: is the wait for ptrace
1113 * @p: task to wait for
1115 * Handle sys_wait4() work for %p in state %TASK_STOPPED or %TASK_TRACED.
1118 * read_lock(&tasklist_lock), which is released if return value is
1119 * non-zero. Also, grabs and releases @p->sighand->siglock.
1122 * 0 if wait condition didn't exist and search for other wait conditions
1123 * should continue. Non-zero return, -errno on failure and @p's pid on
1124 * success, implies that tasklist_lock is released and wait condition
1125 * search should terminate.
1127 static int wait_task_stopped(struct wait_opts
*wo
,
1128 int ptrace
, struct task_struct
*p
)
1130 struct waitid_info
*infop
;
1131 int exit_code
, *p_code
, why
;
1132 uid_t uid
= 0; /* unneeded, required by compiler */
1136 * Traditionally we see ptrace'd stopped tasks regardless of options.
1138 if (!ptrace
&& !(wo
->wo_flags
& WUNTRACED
))
1141 if (!task_stopped_code(p
, ptrace
))
1145 spin_lock_irq(&p
->sighand
->siglock
);
1147 p_code
= task_stopped_code(p
, ptrace
);
1148 if (unlikely(!p_code
))
1151 exit_code
= *p_code
;
1155 if (!unlikely(wo
->wo_flags
& WNOWAIT
))
1158 uid
= from_kuid_munged(current_user_ns(), task_uid(p
));
1160 spin_unlock_irq(&p
->sighand
->siglock
);
1165 * Now we are pretty sure this task is interesting.
1166 * Make sure it doesn't get reaped out from under us while we
1167 * give up the lock and then examine it below. We don't want to
1168 * keep holding onto the tasklist_lock while we call getrusage and
1169 * possibly take page faults for user memory.
1172 pid
= task_pid_vnr(p
);
1173 why
= ptrace
? CLD_TRAPPED
: CLD_STOPPED
;
1174 read_unlock(&tasklist_lock
);
1175 sched_annotate_sleep();
1177 getrusage(p
, RUSAGE_BOTH
, wo
->wo_rusage
);
1180 if (likely(!(wo
->wo_flags
& WNOWAIT
)))
1181 wo
->wo_stat
= (exit_code
<< 8) | 0x7f;
1183 infop
= wo
->wo_info
;
1186 infop
->status
= exit_code
;
1194 * Handle do_wait work for one task in a live, non-stopped state.
1195 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1196 * the lock and this task is uninteresting. If we return nonzero, we have
1197 * released the lock and the system call should return.
1199 static int wait_task_continued(struct wait_opts
*wo
, struct task_struct
*p
)
1201 struct waitid_info
*infop
;
1205 if (!unlikely(wo
->wo_flags
& WCONTINUED
))
1208 if (!(p
->signal
->flags
& SIGNAL_STOP_CONTINUED
))
1211 spin_lock_irq(&p
->sighand
->siglock
);
1212 /* Re-check with the lock held. */
1213 if (!(p
->signal
->flags
& SIGNAL_STOP_CONTINUED
)) {
1214 spin_unlock_irq(&p
->sighand
->siglock
);
1217 if (!unlikely(wo
->wo_flags
& WNOWAIT
))
1218 p
->signal
->flags
&= ~SIGNAL_STOP_CONTINUED
;
1219 uid
= from_kuid_munged(current_user_ns(), task_uid(p
));
1220 spin_unlock_irq(&p
->sighand
->siglock
);
1222 pid
= task_pid_vnr(p
);
1224 read_unlock(&tasklist_lock
);
1225 sched_annotate_sleep();
1227 getrusage(p
, RUSAGE_BOTH
, wo
->wo_rusage
);
1230 infop
= wo
->wo_info
;
1232 wo
->wo_stat
= 0xffff;
1234 infop
->cause
= CLD_CONTINUED
;
1237 infop
->status
= SIGCONT
;
1243 * Consider @p for a wait by @parent.
1245 * -ECHILD should be in ->notask_error before the first call.
1246 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1247 * Returns zero if the search for a child should continue;
1248 * then ->notask_error is 0 if @p is an eligible child,
1251 static int wait_consider_task(struct wait_opts
*wo
, int ptrace
,
1252 struct task_struct
*p
)
1255 * We can race with wait_task_zombie() from another thread.
1256 * Ensure that EXIT_ZOMBIE -> EXIT_DEAD/EXIT_TRACE transition
1257 * can't confuse the checks below.
1259 int exit_state
= READ_ONCE(p
->exit_state
);
1262 if (unlikely(exit_state
== EXIT_DEAD
))
1265 ret
= eligible_child(wo
, ptrace
, p
);
1269 if (unlikely(exit_state
== EXIT_TRACE
)) {
1271 * ptrace == 0 means we are the natural parent. In this case
1272 * we should clear notask_error, debugger will notify us.
1274 if (likely(!ptrace
))
1275 wo
->notask_error
= 0;
1279 if (likely(!ptrace
) && unlikely(p
->ptrace
)) {
1281 * If it is traced by its real parent's group, just pretend
1282 * the caller is ptrace_do_wait() and reap this child if it
1285 * This also hides group stop state from real parent; otherwise
1286 * a single stop can be reported twice as group and ptrace stop.
1287 * If a ptracer wants to distinguish these two events for its
1288 * own children it should create a separate process which takes
1289 * the role of real parent.
1291 if (!ptrace_reparented(p
))
1296 if (exit_state
== EXIT_ZOMBIE
) {
1297 /* we don't reap group leaders with subthreads */
1298 if (!delay_group_leader(p
)) {
1300 * A zombie ptracee is only visible to its ptracer.
1301 * Notification and reaping will be cascaded to the
1302 * real parent when the ptracer detaches.
1304 if (unlikely(ptrace
) || likely(!p
->ptrace
))
1305 return wait_task_zombie(wo
, p
);
1309 * Allow access to stopped/continued state via zombie by
1310 * falling through. Clearing of notask_error is complex.
1314 * If WEXITED is set, notask_error should naturally be
1315 * cleared. If not, subset of WSTOPPED|WCONTINUED is set,
1316 * so, if there are live subthreads, there are events to
1317 * wait for. If all subthreads are dead, it's still safe
1318 * to clear - this function will be called again in finite
1319 * amount time once all the subthreads are released and
1320 * will then return without clearing.
1324 * Stopped state is per-task and thus can't change once the
1325 * target task dies. Only continued and exited can happen.
1326 * Clear notask_error if WCONTINUED | WEXITED.
1328 if (likely(!ptrace
) || (wo
->wo_flags
& (WCONTINUED
| WEXITED
)))
1329 wo
->notask_error
= 0;
1332 * @p is alive and it's gonna stop, continue or exit, so
1333 * there always is something to wait for.
1335 wo
->notask_error
= 0;
1339 * Wait for stopped. Depending on @ptrace, different stopped state
1340 * is used and the two don't interact with each other.
1342 ret
= wait_task_stopped(wo
, ptrace
, p
);
1347 * Wait for continued. There's only one continued state and the
1348 * ptracer can consume it which can confuse the real parent. Don't
1349 * use WCONTINUED from ptracer. You don't need or want it.
1351 return wait_task_continued(wo
, p
);
1355 * Do the work of do_wait() for one thread in the group, @tsk.
1357 * -ECHILD should be in ->notask_error before the first call.
1358 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1359 * Returns zero if the search for a child should continue; then
1360 * ->notask_error is 0 if there were any eligible children,
1363 static int do_wait_thread(struct wait_opts
*wo
, struct task_struct
*tsk
)
1365 struct task_struct
*p
;
1367 list_for_each_entry(p
, &tsk
->children
, sibling
) {
1368 int ret
= wait_consider_task(wo
, 0, p
);
1377 static int ptrace_do_wait(struct wait_opts
*wo
, struct task_struct
*tsk
)
1379 struct task_struct
*p
;
1381 list_for_each_entry(p
, &tsk
->ptraced
, ptrace_entry
) {
1382 int ret
= wait_consider_task(wo
, 1, p
);
1391 static int child_wait_callback(wait_queue_entry_t
*wait
, unsigned mode
,
1392 int sync
, void *key
)
1394 struct wait_opts
*wo
= container_of(wait
, struct wait_opts
,
1396 struct task_struct
*p
= key
;
1398 if (!eligible_pid(wo
, p
))
1401 if ((wo
->wo_flags
& __WNOTHREAD
) && wait
->private != p
->parent
)
1404 return default_wake_function(wait
, mode
, sync
, key
);
1407 void __wake_up_parent(struct task_struct
*p
, struct task_struct
*parent
)
1409 __wake_up_sync_key(&parent
->signal
->wait_chldexit
,
1410 TASK_INTERRUPTIBLE
, p
);
1413 static long do_wait(struct wait_opts
*wo
)
1415 struct task_struct
*tsk
;
1418 trace_sched_process_wait(wo
->wo_pid
);
1420 init_waitqueue_func_entry(&wo
->child_wait
, child_wait_callback
);
1421 wo
->child_wait
.private = current
;
1422 add_wait_queue(¤t
->signal
->wait_chldexit
, &wo
->child_wait
);
1425 * If there is nothing that can match our criteria, just get out.
1426 * We will clear ->notask_error to zero if we see any child that
1427 * might later match our criteria, even if we are not able to reap
1430 wo
->notask_error
= -ECHILD
;
1431 if ((wo
->wo_type
< PIDTYPE_MAX
) &&
1432 (!wo
->wo_pid
|| !pid_has_task(wo
->wo_pid
, wo
->wo_type
)))
1435 set_current_state(TASK_INTERRUPTIBLE
);
1436 read_lock(&tasklist_lock
);
1439 retval
= do_wait_thread(wo
, tsk
);
1443 retval
= ptrace_do_wait(wo
, tsk
);
1447 if (wo
->wo_flags
& __WNOTHREAD
)
1449 } while_each_thread(current
, tsk
);
1450 read_unlock(&tasklist_lock
);
1453 retval
= wo
->notask_error
;
1454 if (!retval
&& !(wo
->wo_flags
& WNOHANG
)) {
1455 retval
= -ERESTARTSYS
;
1456 if (!signal_pending(current
)) {
1462 __set_current_state(TASK_RUNNING
);
1463 remove_wait_queue(¤t
->signal
->wait_chldexit
, &wo
->child_wait
);
1467 static struct pid
*pidfd_get_pid(unsigned int fd
)
1474 return ERR_PTR(-EBADF
);
1476 pid
= pidfd_pid(f
.file
);
1484 static long kernel_waitid(int which
, pid_t upid
, struct waitid_info
*infop
,
1485 int options
, struct rusage
*ru
)
1487 struct wait_opts wo
;
1488 struct pid
*pid
= NULL
;
1492 if (options
& ~(WNOHANG
|WNOWAIT
|WEXITED
|WSTOPPED
|WCONTINUED
|
1493 __WNOTHREAD
|__WCLONE
|__WALL
))
1495 if (!(options
& (WEXITED
|WSTOPPED
|WCONTINUED
)))
1507 pid
= find_get_pid(upid
);
1510 type
= PIDTYPE_PGID
;
1515 pid
= find_get_pid(upid
);
1517 pid
= get_task_pid(current
, PIDTYPE_PGID
);
1524 pid
= pidfd_get_pid(upid
);
1526 return PTR_ERR(pid
);
1534 wo
.wo_flags
= options
;
1543 SYSCALL_DEFINE5(waitid
, int, which
, pid_t
, upid
, struct siginfo __user
*,
1544 infop
, int, options
, struct rusage __user
*, ru
)
1547 struct waitid_info info
= {.status
= 0};
1548 long err
= kernel_waitid(which
, upid
, &info
, options
, ru
? &r
: NULL
);
1554 if (ru
&& copy_to_user(ru
, &r
, sizeof(struct rusage
)))
1560 if (!user_access_begin(infop
, sizeof(*infop
)))
1563 unsafe_put_user(signo
, &infop
->si_signo
, Efault
);
1564 unsafe_put_user(0, &infop
->si_errno
, Efault
);
1565 unsafe_put_user(info
.cause
, &infop
->si_code
, Efault
);
1566 unsafe_put_user(info
.pid
, &infop
->si_pid
, Efault
);
1567 unsafe_put_user(info
.uid
, &infop
->si_uid
, Efault
);
1568 unsafe_put_user(info
.status
, &infop
->si_status
, Efault
);
1576 long kernel_wait4(pid_t upid
, int __user
*stat_addr
, int options
,
1579 struct wait_opts wo
;
1580 struct pid
*pid
= NULL
;
1584 if (options
& ~(WNOHANG
|WUNTRACED
|WCONTINUED
|
1585 __WNOTHREAD
|__WCLONE
|__WALL
))
1588 /* -INT_MIN is not defined */
1589 if (upid
== INT_MIN
)
1594 else if (upid
< 0) {
1595 type
= PIDTYPE_PGID
;
1596 pid
= find_get_pid(-upid
);
1597 } else if (upid
== 0) {
1598 type
= PIDTYPE_PGID
;
1599 pid
= get_task_pid(current
, PIDTYPE_PGID
);
1600 } else /* upid > 0 */ {
1602 pid
= find_get_pid(upid
);
1607 wo
.wo_flags
= options
| WEXITED
;
1613 if (ret
> 0 && stat_addr
&& put_user(wo
.wo_stat
, stat_addr
))
1619 SYSCALL_DEFINE4(wait4
, pid_t
, upid
, int __user
*, stat_addr
,
1620 int, options
, struct rusage __user
*, ru
)
1623 long err
= kernel_wait4(upid
, stat_addr
, options
, ru
? &r
: NULL
);
1626 if (ru
&& copy_to_user(ru
, &r
, sizeof(struct rusage
)))
1632 #ifdef __ARCH_WANT_SYS_WAITPID
1635 * sys_waitpid() remains for compatibility. waitpid() should be
1636 * implemented by calling sys_wait4() from libc.a.
1638 SYSCALL_DEFINE3(waitpid
, pid_t
, pid
, int __user
*, stat_addr
, int, options
)
1640 return kernel_wait4(pid
, stat_addr
, options
, NULL
);
1645 #ifdef CONFIG_COMPAT
1646 COMPAT_SYSCALL_DEFINE4(wait4
,
1648 compat_uint_t __user
*, stat_addr
,
1650 struct compat_rusage __user
*, ru
)
1653 long err
= kernel_wait4(pid
, stat_addr
, options
, ru
? &r
: NULL
);
1655 if (ru
&& put_compat_rusage(&r
, ru
))
1661 COMPAT_SYSCALL_DEFINE5(waitid
,
1662 int, which
, compat_pid_t
, pid
,
1663 struct compat_siginfo __user
*, infop
, int, options
,
1664 struct compat_rusage __user
*, uru
)
1667 struct waitid_info info
= {.status
= 0};
1668 long err
= kernel_waitid(which
, pid
, &info
, options
, uru
? &ru
: NULL
);
1674 /* kernel_waitid() overwrites everything in ru */
1675 if (COMPAT_USE_64BIT_TIME
)
1676 err
= copy_to_user(uru
, &ru
, sizeof(ru
));
1678 err
= put_compat_rusage(&ru
, uru
);
1687 if (!user_access_begin(infop
, sizeof(*infop
)))
1690 unsafe_put_user(signo
, &infop
->si_signo
, Efault
);
1691 unsafe_put_user(0, &infop
->si_errno
, Efault
);
1692 unsafe_put_user(info
.cause
, &infop
->si_code
, Efault
);
1693 unsafe_put_user(info
.pid
, &infop
->si_pid
, Efault
);
1694 unsafe_put_user(info
.uid
, &infop
->si_uid
, Efault
);
1695 unsafe_put_user(info
.status
, &infop
->si_status
, Efault
);
1704 __weak
void abort(void)
1708 /* if that doesn't kill us, halt */
1709 panic("Oops failed to kill thread");
1711 EXPORT_SYMBOL(abort
);